• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

9800GX2 with Multi GPU Folding

lelliott731

[H]ard DCOTM April 2017
Joined
Mar 15, 2002
Messages
941
I was curious as to whether or not anyone out there has a 9800GX2 card, and if so, has tried to fold on both of the GPU's onboard now that the nVidia client now supports multiple GPU folding?

If not, could someone try & post the results of success/failure?
 
Anyone??? I'm really interested in possibly buying one of these, but don't want to waste my $ if the second GPU isn't able to work.
 
A did some google shearches but all the info i got was before the update.

I dunno?

Get from a B&M and return if it's a no go?

 
XP or Vista?

There is a limitation in Vista (thanks to MS) where GPUs must have the desktop extended onto them to be available to CUDA. I'm not sure if you can extend your desktop onto the second half of an X2.

XP doesn't have this limitation, so you might have better luck there.
 
Kendrak, what is a B&M?

bassman, thanks for the info. For now I'm just too afraid to put down that kind of cash till I have heard about someone else who already owned one gets it working.
 
9800 gx2 works fine at least under xp. 4600 ppd on each core at stock and 5000+ with shaders overclocked.

I have mine currently running with a celeron E1400.... 10000ppd box with a <$50 cpu :)
 
Sweet, best bang for the buck I've seen yet.


Hmm, really? In Norway you could get 3x 8800GT's with 512MB for the price of 1x 9800GX2.. Would kick out around 15000PPD+ I'd guess (- overhead).
 
I the UK, Scan is selling the 9800GTX ~£20 cheaper than the 8800GTS.
Its about £80 cheaper now the 280's are out.
The only trouble is they have non of the cheap 9800's in stock, otherwise I'd get 4 ......... :mad:

The 9800GX2 is only 2x the price of a 8800GTS.

Luck ............ :D
 
Awesome, thanks for the info chestRcopRpot2, I am getting some replacement Xeon boards, and they only have 1 PCIe x16 slot, so I want to make as much use of it as I can, and that's the cheapest way to get 10k PPD boost out of one PCIe slot.
 
Awesome, thanks for the info chestRcopRpot2, I am getting some replacement Xeon boards, and they only have 1 PCIe x16 slot, so I want to make as much use of it as I can, and that's the cheapest way to get 10k PPD boost out of one PCIe slot.

Which boards you going for ??
Anybody seen a review of the Asus Z7S WS ?
Link to homepage

Luck ......... :D
 
That was one of the ones I was looking for. I'm looking from now on for boards with Power/Reset buttons on them, that way I can stop buying cases, and can arrange airflow with large fans in my computer room.

Now I need to find foam padding for the walls to block sound, and strips of foam around the doors so it doesn't leak out. That way my wife won't notice if I add another machine or two in a month or so so I don't get completely wiped by Xilikon.
 
Which boards you going for ??
Anybody seen a review of the Asus Z7S WS ?
Link to homepage

Luck ......... :D

The Z7S has had some serious bios issues, they seem to have fixed the snoop filter problem though, but I'd still be leary. If you're really going to buy a 5400 series board with a gimped memory layout, you're best off going with a D5400XS and you'll get an extra PCIe slot out of it too.
 
My D5400XS has the power & reset buttons on the mobo.

I've got my 5450's up to 3.4 Ghz (378x9) stable on stock voltage so far.
They will boot at 400x8 but the 400 FSB is not stable.
Either the memory or NB is running to hot and it BSOD's on me at that FSB speed.
I've active cooling on both.

The best thing I like about this mobo vs the other Xeon mobo I got is that it takes S775 heatsinks and theres enough room for a pair of TRUE's to fit with push/pull fans.

The one main down side I've got bar the heat is I've got my vid card in the 3rd PCI-e slot.
To reset the bios if I overclock it to far means removing the vid card to get at the jumper.

One other thing, theres no memory multi, so get DDR2-800 FB Dimms if you want to run adove 333 FSB.

Luck ........... :D
 
The D5400XS is the board that has the nVidia N100 chips on it right? So it can run nVidia SLI? Which means I could load my boards up with my current Xeons and 2 or more 8800GTS 512MB cards, or 9800GX2's if I can move enough air and have enough $.
 
You can run two cards in SLI on a D5400XS Skulltrails.
Tri SLI is not supported.

There are 4x PCI-e x16 slots but the bottom two are only single width.
So your only looking to fit 3x double width cards into the mobo in reality.

I like the mobo.
Its stable even with my E5450's overclocked to 3.4 Ghz.
Its not that bad power wise.
It only pulls in around 300 watts plus what the vid card draws.

But it is expensive to set up.

Luck .......... :D
 
Right, but if I didn't care to do SLI, and I only wanted to build the fastest possible PPD machine, that board would enable me to have 3 9800GX2's on board, and overclock the Xeons right?
 
Right, but if I didn't care to do SLI, and I only wanted to build the fastest possible PPD machine, that board would enable me to have 3 9800GX2's on board, and overclock the Xeons right?

Yes.

But the MSI K9A2 Platinum mobo is a better bet for a maximum PpD box.
4x 9800GX2's vs 3x 9800GX2's
A single Phenom is a lot cheaper than 2x Xeons. The money you'd save on the CPU's would easy get you the fourth vid card.
You only need DDR2 Dimms not DDR2-FB Dimms, runs a lot cooler.
But you'd have to run Vista on it so you could make use of all the GPU cores.

Luck .......... :D
 
9800 gx2 works fine at least under xp. 4600 ppd on each core at stock and 5000+ with shaders overclocked.

I have mine currently running with a celeron E1400.... 10000ppd box with a <$50 cpu :)
Shouldn't each GPU on the GX2 card produce well over 5000PPD each? Aren't there 128 stream processors per GPU, as much as the 9800GTX? My OC 8800GT is producing 5000PPD. I would expect closer to 6000PPD per GPU on a GX2 with OC shaders, unless I'm entirely missing something.
 
Shouldn't each GPU on the GX2 card produce well over 5000PPD each? Aren't there 128 stream processors per GPU, as much as the 9800GTX? My OC 8800GT is producing 5000PPD. I would expect closer to 6000PPD per GPU on a GX2 with OC shaders, unless I'm entirely missing something.
If I remember rightly it's 2x 8800GTS 512 so I think 5500 PPD per core should be achievable with a hearty OC. :)

It was a toss up for me between the GX2 and the 9800GTX. I actually had a GX2 on pre-order but came to my senses and went for the GTX. Can't afford that outlay for folding right at the minute. Would have been very cool though.

 
Shouldn't each GPU on the GX2 card produce well over 5000PPD each? Aren't there 128 stream processors per GPU, as much as the 9800GTX? My OC 8800GT is producing 5000PPD. I would expect closer to 6000PPD per GPU on a GX2 with OC shaders, unless I'm entirely missing something.

Heat.

The GX2 is shifting almost twice the heat of a GTX in the same space.
Which is why the GX2 starts with a lower stock clock then the GTX.
With the 4 cards stacked together, it then depends on how much heat you can disipate, if you can overclock them at all.

Luck ............ :D
 
Heat.

The GX2 is shifting almost twice the heat of a GTX in the same space.
Which is why the GX2 starts with a lower stock clock then the GTX.
With the 4 cards stacked together, it then depends on how much heat you can disipate, if you can overclock them at all.

Luck ............ :D

most likely your best bet would be to under clock the GPU abit and memory...

Say if stock is 600/1700/1800 I'd drop it down to like 550/1900+/1600... That will produce just as much or more than if you run yourself up to heat limits by OC'n gpu and mem!

Trust me my GTS which I under clocked about 50mhz and hit 2000 on the shaders has been purring at nearly 6k PPD :eek:
 
most likely your best bet would be to under clock the GPU abit and memory...

Say if stock is 600/1700/1800 I'd drop it down to like 550/1900+/1600... That will produce just as much or more than if you run yourself up to heat limits by OC'n gpu and mem!

Trust me my GTS which I under clocked about 50mhz and hit 2000 on the shaders has been purring at nearly 6k PPD :eek:
Doesn't work for me. No matter how low the core and memory frequencies are reduced, I can hardly increase the shaders beyond what I have them clocked at now. Moreover, if I decrease the other clocks too far, performance suffers and it becomes self defeating. I only have single slot cards, so this method may work with dual slot cards or aftermarket cooled cards.

BTW, anyone seen aftermarket coolers for the GX2 cards excluding WC setups?
 
That is really weird! shame really, but then again the total output still crushes any single gpu solution out there! :)

 
That is really weird! shame really, but then again the total output still crushes any single gpu solution out there! :)
You mean single-slot solution? Yes it's true, but in my case it was out of necessity rather than choice. My motherboard doesn't have many free slots available that are adjacent to the 16x PCI-E slots, and I don't even have a sound card installed (using sucky onboard sound). :rolleyes:

This board has a plethora of different slot types (PCI, PCI-X, 4X and 16X PCI-E) which has proved limiting. I wanted to move my SSD card into a PCI-X slot, but it's only compatible with standard 32-bit PCI, otherwise I'd have chosen dual-slot GTS cards which produce at least ~500PPD more per card compared to the GT cards for a little higher cost.

If I opt to get rid of the SSD card altogether, I could liberate an adjacent slot for one GX2 card and keep a single 8800GT, but the potential issues of mismatching cards and the consequent performance plummet has me concerned. Anyone know if there's a significant difference between a GX2 and GT to warrant avoiding mixing them for folding?

 
If I opt to get rid of the SSD card altogether, I could liberate an adjacent slot for one GX2 card and keep a single 8800GT, but the potential issues of mismatching cards and the consequent performance plummet has me concerned. Anyone know if there's a significant difference between a GX2 and GT to warrant avoiding mixing them for folding?


I'd think it would be an issue since the GT is 112 SP vs the 256 on the GX2. But there is only one way to know for sure how bad it would be. Then decide if it's worth it. The GT may still pull 4000-4200ppd and if not below that I'd say it may be worth it. I guess only you can really make that call.

Luck!

 
I'd think it would be an issue since the GT is 112 SP vs the 256 on the GX2. But there is only one way to know for sure how bad it would be. Then decide if it's worth it. The GT may still pull 4000-4200ppd and if not below that I'd say it may be worth it. I guess only you can really make that call.
I guess up to 1000PPD drop on the single GT is tolerable, but wouldn't the GPUs on the GX2 also suffer a loss in performance? Is Stanford going to fix this issue or is it something we will always need to consider with future video card purchases?
 
I guess up to 1000PPD drop on the single GT is tolerable, but wouldn't the GPUs on the GX2 also suffer a loss in performance? Is Stanford going to fix this issue or is it something we will always need to consider with future video card purchases?

The primary card (1st slot) shouldn't suffer, only the second card does. Far as I know it's not a Stanford/FAH issue. It appears to be more of a driver or maybe Cuda or chipset/hardware limitation. Not very specific there I know, but these are what I've been able to gather so far.
If you don't already have the GT to test with maybe there's someone close to you that has one you could test with?

 
Is there a similar limitation on mismatched multiple GPUs with ATI cards? If they aren't then we would know it was related to the drivers or CUDA.
 
The primary card (1st slot) shouldn't suffer, only the second card does. Far as I know it's not a Stanford/FAH issue. It appears to be more of a driver or maybe Cuda or chipset/hardware limitation. Not very specific there I know, but these are what I've been able to gather so far.
If you don't already have the GT to test with maybe there's someone close to you that has one you could test with?

In my experience either card can suffer depending on which client you start first, I've had the low PPD on either of my cards at some point, with the other going full speed.

 
flogge said:
Is there a similar limitation on mismatched multiple GPUs with ATI cards? If they aren't then we would know it was related to the drivers or CUDA.

I'm not sure about the ATI cards. Maybe someone running dual ATI cards can chime in here.

Dan-515 said:
In my experience either card can suffer depending on which client you start first, I've had the low PPD on either of my cards at some point, with the other going full speed.

That's interesting. I've never tried starting the second card first. I may have to try that out and see if it works that way on my system.

 
Back
Top